基于BP神经网络对黄河源区降水的降尺度模拟研究  

Downscaling Analysis of Precipitation in Source Region of Yellow River Simulated by Using BP Neural Network

作　　者：石晓燕[1,2] 郑庆康[2] 杨涛[1] 

机构地区：[1]河海大学水文水资源与水利工程科学国家重点实验室,南京210098 [2]河海大学岩土工程科学研究所,南京210098

出　　处：《三峡大学学报（自然科学版）》2013年第5期36-40,共5页Journal of China Three Gorges University：Natural Sciences

基　　金：国家自然科学基金资助项目(40901016;40830639);国家重点基础发展研究(973)资助项目(2010CB428405);水文水资源与水利工程科学国家重点实验室自主探索课题资助项目(2009586612;2009585512)

摘　　要：利用黄河源区实测的水文气象资料、NCEP再分析数据及HadCM3情景模式,通过研究气象因子与降雨之间的非线性关系,基于BP神经网络模型采用3种不同时段(即全年S1、枯水期和丰水期S2及四季S4)对黄河源区建立3种降雨的统计降尺度模式,分析评价不同模式对降雨指标的模拟能力.结果表明:3个模型的模拟效果的优劣随着研究的时间尺度变化,对多数降雨指标的变化趋势能成功捕捉,但对量的模拟效果会随着指标和季节而变化,对极端指标的模拟效果均较差.The nonlinear relationships between meteorological factors and precipitation are investigated; and the downscaling analysis of precipitation in the source region of the Yellow River are simulated by using BP neural network model. Three methods are applied to study the downscaling analysis of precipitation; and three different statistical downscaling models are established. According to the analysis and assessment re- sults about the simulation effect of precipitation indexes applying different models, it is indicated that the sim- ulation ability of three methods varies with time scale. The trend of most precipitation indices can be simula- ted reasonably well. However, the quantitative simulation results vary with the indices and seasons, resulting in a worse simulation effect for extreme indices.

关 键 词：BP神经网络 统计降尺度法 气候因子 降水预测 

分 类 号：TV223[水利工程—水工结构工程] TV214